Hi Peter,
Peter Tribble wrote:
> I added a jfsstat tool to my jkstat package:
>
> http://www.petertribble.co.uk/Solaris/jkstat.html
Nice work!
> This was quite easy - once I had decided not to try and show all
> 45 (or so) statistics in a horizontal line of the table all at
> once.
Having a menu to select which fields one is interested in
might be useful. But there are certainly several stats (e.g.,
nvnevent) which would probably never, ever get looked at.
> It's very neat being able to see what's going on!
>
> And I implemented sorting on the table too. That's a nice feature,
> something generally missing from many tools.
Agreed. Something like that is being considered for 'fsstat'.
> I also filter out the aggregate (ie. ufs, zfs, nfs) statistics.
>
> I'm not at all convinced that having the aggregate statistics
> helps (and I wonder if there's a performance hit). Normally,
> I'm interested in per-filesystem activity - so the aggregates
> aren't useful. If I want them, I can get them by aggregating
> the filesystem statistics. (This might differ from the aggregate
> if things have been unmounted, but in that case you need to know
> the full history to make sense of the numbers.)
>
> Presumably the per-fstype statistics are there for a reason?
> Is there one, because nothing is immediately obvious.
Aside from general curiousity, there is one useful aspect
of seeing statistics by fstype: NFS. It's useful to see
which version of NFS is getting the most use on a client.
Okay, okay... I'll admit that this is of more interest
to an NFS developer... In fact, I'll go so far as to say
the fstype statistics are probably more interesting to
FS developers than anyone else. :-)
> I also tried to filter per-zone (by parsing /etc/mnttab). This
> worked less well than I expected, as:
> - lofs entries in /etc/mnttab don't have a zone flag
> - the lofs statistics seem only partial - most things seem to
> go right through to the underlying filesystem.
>
> Again, I wonder if the lofs statistics are worthwhile?
>
> (I haven't actually tried zones on my test system yet. I'm
> still working on this...)
>
With regard to zones and fsstat kstats: I've been talking with
Dan and the zones folks about how to make fsstat work better
with zones. We have a few ideas (as Dan mentioned) but the
upshot is that the format of the kstat names *may* change
to support zones.
Keep up the great work!
Rich
